NameError: name 'init_empty_weights' is not defined

[icode]

System Info

python 3.13
M4 macos 15.4

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

do just

if torch.cuda.is_available():
    device = torch.device("cuda")
else:
    device = torch.device("cpu")
model = CLIPModel.from_pretrained(MODEL_NAME).to(device)

Expected behavior

Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Traceback (most recent call last):

  File "/Users/icode/Projects/self/fanfan/ai-backend/app.py", line 5, in <module>

    import handlers

  File "/Users/icode/Projects/self/fanfan/ai-backend/handlers.py", line 16, in <module>

    model = CLIPModel.from_pretrained(MODEL_NAME).to(device)

            ~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^

  File "/Users/icode/Projects/self/fanfan/ai-backend/.venv/lib/python3.13/site-packages/transformers/modeling_utils.py", line 279, in _wrapper

    return func(*args, **kwargs)

  File "/Users/icode/Projects/self/fanfan/ai-backend/.venv/lib/python3.13/site-packages/transformers/modeling_utils.py", line 4333, in from_pretrained

    model_init_context = cls.get_init_context(is_quantized, _is_ds_init_called)

  File "/Users/icode/Projects/self/fanfan/ai-backend/.venv/lib/python3.13/site-packages/transformers/modeling_utils.py", line 3736, in get_init_context

    init_contexts = [no_init_weights(), init_empty_weights()]

                                        ^^^^^^^^^^^^^^^^^^

NameError: name 'init_empty_weights' is not defined

[zbowling]

same issue. regression in transformers

[Ahajha]

Caused by #37306.

[zbowling]

Maybe fixed by #37324

[Samoed]

Same issue #37311

[Vishesh-Mistry]

I was facing the same issue, and downgrading to transformers==4.50.3 resolved it for me.
I found this solution by referring to #37311 (comment)

[Ahajha]

You can also add accelerate as a dependency. This issue pops up in the first place because init_empty_weights is from that library, and is only conditionally imported (if it exists).

[harupy]

Ran into the same error. Can 4.51.0 be yanked?

[ArthurZucker]

we will patch today

[ArthurZucker]

sorry everyone for this! cc @Cyrilvallez

[Cyrilvallez]

This will be fixed in #37337! In the meantime, you can run pip install accelerate to solve it! Super sorry for the disturbance!

[ArthurZucker]

Closing as this was fixed on main #37337 patching asap

[Ahajha]

Can we get a patch release for this?

[RoyceAroc]

whats the eta on the patch

[ArthurZucker]

1h max

[ArthurZucker]

Patch is out for 3 hours sorry for not mentionning !